AI-driven robotics key to recycling’s challenges

By admin In News, Technology No comments

AI-driven robotics key to recycling’s challenges

China’s Operation National Sword policy initiative curbed the country’s imports of most types of solid waste for disposal, and imposed stringent limits on the materials that it will accept for recycling. The policy’s ramifications have hugely impacted the many foreign economies that for decades relied on China’s materials recovery facilities (MRFs) to deal with the bulk of their recyclable waste materials, from plastics and packaging to glass, metals, and wood.

Before National Sword came into force in 2018, 70 per cent of plastics collected in the US and 95 per cent of plastics collected in the European Union (EU) were consigned to a slow boat to China, according to the Yale School of the Environment. A study by the University of Georgia estimated that by 2030, the policy could have ‘displaced’ some 111 million tonnes of discarded plastic.

As other eastern nations also impose strict limits on the amount of western refuse they will accept, the waste disposal problem exporters now face has caused many to fundamentally reconsider their collection, recovery and processing strategies for recyclables. In particular, it has highlighted the essential importance of ensuring that waste for recycling is as ‘pure’ as possible – at least 98 per cent in some instances. This means uncontaminated by use residues (such as food deposits), unmixed with non-recyclables, and reliably sorted into same-material silos.

To this end, MRFs in Europe and North America are investing in a new-generation scalable waste-processing plant. It will combine vision-assisted robotics with cloud-based artificial intelligence (AI) back ends to fully automate the way mixed waste is processed for recycling. The last year has seen a wave of new solutions installed and enter operation.

In November 2020, Texas-based waste services specialist Waste Connections announced the deployment of 24 AI-guided robotics solutions from AMP Robotics. Norwegian waste-management firm Bjorstaddalen opened the country’s first municipal AI-powered robotic waste-sorting station in April 2021, running on technology from ZenRobotics.

In the UK, Coventry City Council has contracted another AI waste-management solutions provider, Machinex, to deliver an AI-backed MRF that will be able to process some 175,000 tonnes of recyclables per year. Operated by Sherbourne Recycling, the MRF will use 14 Machinex SamurAI sorting robots and 14 optical sorters.

In France, Machinex has sold six more SamurAIs to the Veolia Group as MRFs located at Portes-lès-Valence and Bègles are upgraded.
In November 2020, Texas-based waste services specialist Waste Connections announced the deployment of 24 AI-guided robotics solutions from AMP Robotics.

Integrated into existing MRF plants as modular units, this emergent technology class employs advanced vision systems that use machine intelligence to recognise broad ranges of waste types as they are conveyed through the facility. The various waste types are then crosschecked against datasets, which identify their material and other characteristics necessary for recycling.

However, before suitable waste is near ready to become reusable, some pre-processes must occur; it has to be picked and sorted. These processes are necessary if the recovered materials are to be turned into the raw materials that can be used to make new recyclable objects, which will help turn the economic circularity many countries aim for.

For this, robotic arms are guided to separate waste fractions from the conveyor, and put them into assigned collection chutes for aggregation. The arms work fast, at rates in the range of 2,000-4,200 picks-per-hour-​per-sorting bay (depending on use requirements), compared to a typical manual (or human) waste-picker rate of 200 picks per hour.

“Traditionally, waste processors relied on manual pickers to identify and sort waste,” says Peter Hedley, chief technology officer at waste management start-up Recycleye. “They are exposed to a multitude of occupational hazards, which has resulted in the waste-recovery industry facing an average of 50 per cent labour turnover every six months.”

‘The digitisation of the flow of recyclables provides the ability to measure material streams… Materials recovery facility managers can use this data to make more-informed decisions.’

Matanya Horowitz, AMP Robotics

 

Compounded by China’s Operation National Sword and coronavirus workplace rules, the waste recovery industry is now overstretched.
Furthermore, most waste facilities use material recovery modules that were developed up to 20 years ago. Hedley says: “Due to the changing waste composition since then, the systems are forced to operate at a reduced capacity due to the inability to effectively sort more abundant materials.”

Robotic picker arms work faster and longer than their human counterparts, and can lift heavier objects. Depending on the waste type involved, robotic arms can hold and lift objects that weigh anything between 1.25kg and 30kg, according to the model capacity and gripper type (e.g. air suction or mechanical claw).

Some solutions can be configured for secondary applications, such as quality control (removing unwanted items from a single waste stream that’s mostly pure) and sucking out unrecyclable materials like plastic bags and film, which can jam up plant machinery and cause stoppages.

Growing interest in the intelligent waste-management market has opened opportunities for robotic automation specialists to join with AI specialists to create integrated modular solutions. One such collaboration is factory automation specialist FANUC UK, which in April 2021 partnered with Recycleye to create a robotic waste-picking system powered by cloud-based AI that runs on Microsoft Azure.

“At the start of a new installation the vision system is installed in the MRF,” says Recycleye’s Hedley. “This detects the waste passing on [the existing] conveyor belts. This creates millions of images of waste, millions of data points – data points that we use to teach our AI models what waste looks like. When the system goes into full operation our datasets grow bigger and bigger, and Azure scales to get as big as our datasets are.”

The AI component of intelligent waste-management applications also feed analytics dashboards than run behind the materials management operations. These capture process data that, as datasets build up, provide MRF operators with more detailed understanding of waste dynamics. AI also informs an MRF’s ability to offer value-added services to clients at either end of the waste-processing supply chain.

Suppliers of waste can gain insights into what their stuff contains, and data can be fed back up the waste chain to improve recyclability at the points of collection. While many waste materials are not suitable to be recycled, some waste fractions that can be recycled have negligible resale value, and to process them into recycled raw materials could be environmentally detrimental.

AI technology’s ability to reconfigure waste sorting patterns in response to materials market demand fluctuations is another way for MRFs to add value to their operations, especially in situations where specific recyclables are ultimately more likely to be bought. Key to this is the AI’s ability to include new material types in its datasets, and to recognise valued materials when they are in a dirty or damaged condition, says Matanya Horowitz, founder and CEO at AMP Robotics.

“AMP Robotics’ AI has to learn many different material types, and deformations – as well as customer-specific material types, which vary in terms of end-market application,” he explains. “AI enables the digitisation of objects in the material stream, which opens up many applications as this technology matures. The digitisation of the flow of recyclables provides the ability to measure material streams, and then our AMP Clarity portal surfaces the data and insights that MRF managers can use to make more-informed decisions.”
Horowitz adds: “For example, they can better understand how much contamination enters the facility (and from what sources), the loss of valuable materials at the point-of-residue, and whether they can drive a higher price for scrap bales based on both the quantity and the quality of material.”

Robotic sorting, along with the emergence of descriptive and diagnostic analysis through stand-alone sensors, should be seen as initial applications for AI-enabled waste-management technology, according to Horowitz.

“The distribution of these sensors throughout MRFs helps them become more data-driven,” he points out. “Traditionally, MRFs are centralised material hubs, but the proliferation of these sensors begins to effectively transform recovery facilities into digital information hubs.”

E-waste innovation

‘Disassembly line’ enables valued fraction recovery for the first time

Brand-new electronic goods like flatscreen TVs and computer displays present recyclers with a longstanding dilemma. They contain valued materials that can be recovered and reused. They also contain varying amounts of noxious substances that, released into the immediate vicinity by the recycling processes, could prove harmful to MRF personnel and to organisms further afield.

Until recently, the process for recycling flat-​panel displays had not changed since the start of the Waste from Electrical and Electronic Equipment (WEEE) Regulations in 2006, says Craig Thompson, CEO at UK screen-based-device recycling specialist Areera.

“Shredding [has been] the main process for recycling flat-panel displays, with hazardous chemical particles – compounds of arsenic, cadmium, lead, mercury – covering the shredder and the recovery facilities,” he says. “The shredded raw materials themselves – ferrous, nonferrous and circuit boards – also cause potential hazards on the downstream processing of the materials.”

Shredding yields minimal recoverable value – there’s almost no market for co-mingled metal, plastics, and glass – and it’s a power-hungry process that does not comply with current low-energy/low-emission standards.

Areera is tackling the problem with newly installed solutions from robotic recycling system provider FPD Recycling. Its FPD PRO ‘depollution machines’ integrate AI with two ABB Robotics model IRB 2600 20kg-payload robotic arms – each fitted with additional cutting unit axes – that at full capacity can dissemble and process up to 100 screens or 200 laptop PCs per hour.

Processing happens in three stations. Defunct screen-based devices are placed vertically on the in-feed conveyor belt, and moved through the container. In the first station, the object is weighed. Data is collected and the object is identified and categorised.

It then gets conveyed to the next station, where it is held in position; the first robot arm scans the object’s screen and determines the processing method, and then removes the screen.

If the screen is identified as a Cold Cathode Fluorescent Lamp (CCFL) TV, for instance, it is conveyed to the next station where the second robot takes an image to identify the location of CCFLs, and then removes the mercury hazards from the chassis. The depolluted display then exits the station safe for the next process step.

The scanner also looks for a manufacturer logo if it’s visible on the front panel/bezel. All the information is fed into a continually building dataset that informs future object identification. In the next implementation of FPD PRO, a barcode scanner will look for the label and record information included there – barcode/QR Code and model number, for example.

“Obtaining specifics on any plastics component is important principally because many products combine polymers with a bromine flame retardant, which is now deemed to be hazardous over a given concentration,” explains Paudy O’Brien, CEO at FPD Recycling. “Identification and separation mean that the bromine-free fraction can be recovered.”

Waste recognition

Branding to improve AI logo detection

Despite the superabundance of mixed waste that potentially could be recycled, gaining knowledge of its key constituent statistics is an estimable challenge. “The largest challenge for AI is the sheer variety of waste,” says Peter Hedley at Recycleye.

“Waste differs by country, for instance, and as waste objects can get crushed into strange shapes, get covered in dirt or discolour. AI systems must learn to see past these eventualities to identify the item consistently.”

If initiatives such as Extended Producer Responsibility (a framework that would enfold the environmental costs associated with a product throughout its life cycle into its market price) are to succeed, they need greater transparency as to who – which producer – made it.

A development of EPR that AI-enabled waste management is investigating is in encouraging producers to design packaging so that it is more AI-friendly, i.e. by developing designs that makes it easier for vision systems to recognise specific objects for their recyclability.

Such initiatives are at early stages. Both Recycleye and AMP Robotics are trialling AI/computer vision for brand-level logo detection of retail waste in real-time operation. Recycleye’s algorithm was trained with plastic drinks brands Coca Cola, Pepsi, and Heineken, using 1,399 images from the internet.

An overall accuracy of 93 per cent was obtained with varying lighting conditions, damaged and partially visible objects. Recycleye reports: “With improvement, this technology can be implemented [in MRFs] to support a closed-loop system of re-use. It also provides an Extended Producer Responsibility [for] brands [to] take greater accountability for the environmental impacts of their product-end-life.”

AMP Robotics is working with consumer packaged goods companies to recover higher rates of these materials used in some of their products. One of AMP Robotics’ first initiative partners is beverage brand Keurig Dr Pepper (KDP) in support of its introduction of recyclable K-Cup pods. Following KDP’s conversion of its coffee pods to polypropylene – a valued material for recycled plastics – the companies trained AMP Robotics’ technology to more exactly identify and sort K-Cup pods, with the aim of increasing waste recovery levels.

China’s Operation National Sword policy initiative curbed the country’s imports of most types of solid waste for disposal, and imposed stringent limits on the materials that it will accept for recycling. The policy’s ramifications have hugely impacted the many foreign economies that for decades relied on China’s materials recovery facilities (MRFs) to deal with the bulk of their recyclable waste materials, from plastics and packaging to glass, metals, and wood.

Before National Sword came into force in 2018, 70 per cent of plastics collected in the US and 95 per cent of plastics collected in the European Union (EU) were consigned to a slow boat to China, according to the Yale School of the Environment. A study by the University of Georgia estimated that by 2030, the policy could have ‘displaced’ some 111 million tonnes of discarded plastic.

As other eastern nations also impose strict limits on the amount of western refuse they will accept, the waste disposal problem exporters now face has caused many to fundamentally reconsider their collection, recovery and processing strategies for recyclables. In particular, it has highlighted the essential importance of ensuring that waste for recycling is as ‘pure’ as possible – at least 98 per cent in some instances. This means uncontaminated by use residues (such as food deposits), unmixed with non-recyclables, and reliably sorted into same-material silos.

To this end, MRFs in Europe and North America are investing in a new-generation scalable waste-processing plant. It will combine vision-assisted robotics with cloud-based artificial intelligence (AI) back ends to fully automate the way mixed waste is processed for recycling. The last year has seen a wave of new solutions installed and enter operation.

In November 2020, Texas-based waste services specialist Waste Connections announced the deployment of 24 AI-guided robotics solutions from AMP Robotics. Norwegian waste-management firm Bjorstaddalen opened the country’s first municipal AI-powered robotic waste-sorting station in April 2021, running on technology from ZenRobotics.

In the UK, Coventry City Council has contracted another AI waste-management solutions provider, Machinex, to deliver an AI-backed MRF that will be able to process some 175,000 tonnes of recyclables per year. Operated by Sherbourne Recycling, the MRF will use 14 Machinex SamurAI sorting robots and 14 optical sorters.

In France, Machinex has sold six more SamurAIs to the Veolia Group as MRFs located at Portes-lès-Valence and Bègles are upgraded.
In November 2020, Texas-based waste services specialist Waste Connections announced the deployment of 24 AI-guided robotics solutions from AMP Robotics.

Integrated into existing MRF plants as modular units, this emergent technology class employs advanced vision systems that use machine intelligence to recognise broad ranges of waste types as they are conveyed through the facility. The various waste types are then crosschecked against datasets, which identify their material and other characteristics necessary for recycling.

However, before suitable waste is near ready to become reusable, some pre-processes must occur; it has to be picked and sorted. These processes are necessary if the recovered materials are to be turned into the raw materials that can be used to make new recyclable objects, which will help turn the economic circularity many countries aim for.

For this, robotic arms are guided to separate waste fractions from the conveyor, and put them into assigned collection chutes for aggregation. The arms work fast, at rates in the range of 2,000-4,200 picks-per-hour-​per-sorting bay (depending on use requirements), compared to a typical manual (or human) waste-picker rate of 200 picks per hour.

“Traditionally, waste processors relied on manual pickers to identify and sort waste,” says Peter Hedley, chief technology officer at waste management start-up Recycleye. “They are exposed to a multitude of occupational hazards, which has resulted in the waste-recovery industry facing an average of 50 per cent labour turnover every six months.”

‘The digitisation of the flow of recyclables provides the ability to measure material streams… Materials recovery facility managers can use this data to make more-informed decisions.’

Matanya Horowitz, AMP Robotics

 

Compounded by China’s Operation National Sword and coronavirus workplace rules, the waste recovery industry is now overstretched.
Furthermore, most waste facilities use material recovery modules that were developed up to 20 years ago. Hedley says: “Due to the changing waste composition since then, the systems are forced to operate at a reduced capacity due to the inability to effectively sort more abundant materials.”

Robotic picker arms work faster and longer than their human counterparts, and can lift heavier objects. Depending on the waste type involved, robotic arms can hold and lift objects that weigh anything between 1.25kg and 30kg, according to the model capacity and gripper type (e.g. air suction or mechanical claw).

Some solutions can be configured for secondary applications, such as quality control (removing unwanted items from a single waste stream that’s mostly pure) and sucking out unrecyclable materials like plastic bags and film, which can jam up plant machinery and cause stoppages.

Growing interest in the intelligent waste-management market has opened opportunities for robotic automation specialists to join with AI specialists to create integrated modular solutions. One such collaboration is factory automation specialist FANUC UK, which in April 2021 partnered with Recycleye to create a robotic waste-picking system powered by cloud-based AI that runs on Microsoft Azure.

“At the start of a new installation the vision system is installed in the MRF,” says Recycleye’s Hedley. “This detects the waste passing on [the existing] conveyor belts. This creates millions of images of waste, millions of data points – data points that we use to teach our AI models what waste looks like. When the system goes into full operation our datasets grow bigger and bigger, and Azure scales to get as big as our datasets are.”

The AI component of intelligent waste-management applications also feed analytics dashboards than run behind the materials management operations. These capture process data that, as datasets build up, provide MRF operators with more detailed understanding of waste dynamics. AI also informs an MRF’s ability to offer value-added services to clients at either end of the waste-processing supply chain.

Suppliers of waste can gain insights into what their stuff contains, and data can be fed back up the waste chain to improve recyclability at the points of collection. While many waste materials are not suitable to be recycled, some waste fractions that can be recycled have negligible resale value, and to process them into recycled raw materials could be environmentally detrimental.

AI technology’s ability to reconfigure waste sorting patterns in response to materials market demand fluctuations is another way for MRFs to add value to their operations, especially in situations where specific recyclables are ultimately more likely to be bought. Key to this is the AI’s ability to include new material types in its datasets, and to recognise valued materials when they are in a dirty or damaged condition, says Matanya Horowitz, founder and CEO at AMP Robotics.

“AMP Robotics’ AI has to learn many different material types, and deformations – as well as customer-specific material types, which vary in terms of end-market application,” he explains. “AI enables the digitisation of objects in the material stream, which opens up many applications as this technology matures. The digitisation of the flow of recyclables provides the ability to measure material streams, and then our AMP Clarity portal surfaces the data and insights that MRF managers can use to make more-informed decisions.”
Horowitz adds: “For example, they can better understand how much contamination enters the facility (and from what sources), the loss of valuable materials at the point-of-residue, and whether they can drive a higher price for scrap bales based on both the quantity and the quality of material.”

Robotic sorting, along with the emergence of descriptive and diagnostic analysis through stand-alone sensors, should be seen as initial applications for AI-enabled waste-management technology, according to Horowitz.

“The distribution of these sensors throughout MRFs helps them become more data-driven,” he points out. “Traditionally, MRFs are centralised material hubs, but the proliferation of these sensors begins to effectively transform recovery facilities into digital information hubs.”

E-waste innovation

‘Disassembly line’ enables valued fraction recovery for the first time

Brand-new electronic goods like flatscreen TVs and computer displays present recyclers with a longstanding dilemma. They contain valued materials that can be recovered and reused. They also contain varying amounts of noxious substances that, released into the immediate vicinity by the recycling processes, could prove harmful to MRF personnel and to organisms further afield.

Until recently, the process for recycling flat-​panel displays had not changed since the start of the Waste from Electrical and Electronic Equipment (WEEE) Regulations in 2006, says Craig Thompson, CEO at UK screen-based-device recycling specialist Areera.

“Shredding [has been] the main process for recycling flat-panel displays, with hazardous chemical particles – compounds of arsenic, cadmium, lead, mercury – covering the shredder and the recovery facilities,” he says. “The shredded raw materials themselves – ferrous, nonferrous and circuit boards – also cause potential hazards on the downstream processing of the materials.”

Shredding yields minimal recoverable value – there’s almost no market for co-mingled metal, plastics, and glass – and it’s a power-hungry process that does not comply with current low-energy/low-emission standards.

Areera is tackling the problem with newly installed solutions from robotic recycling system provider FPD Recycling. Its FPD PRO ‘depollution machines’ integrate AI with two ABB Robotics model IRB 2600 20kg-payload robotic arms – each fitted with additional cutting unit axes – that at full capacity can dissemble and process up to 100 screens or 200 laptop PCs per hour.

Processing happens in three stations. Defunct screen-based devices are placed vertically on the in-feed conveyor belt, and moved through the container. In the first station, the object is weighed. Data is collected and the object is identified and categorised.

It then gets conveyed to the next station, where it is held in position; the first robot arm scans the object’s screen and determines the processing method, and then removes the screen.

If the screen is identified as a Cold Cathode Fluorescent Lamp (CCFL) TV, for instance, it is conveyed to the next station where the second robot takes an image to identify the location of CCFLs, and then removes the mercury hazards from the chassis. The depolluted display then exits the station safe for the next process step.

The scanner also looks for a manufacturer logo if it’s visible on the front panel/bezel. All the information is fed into a continually building dataset that informs future object identification. In the next implementation of FPD PRO, a barcode scanner will look for the label and record information included there – barcode/QR Code and model number, for example.

“Obtaining specifics on any plastics component is important principally because many products combine polymers with a bromine flame retardant, which is now deemed to be hazardous over a given concentration,” explains Paudy O’Brien, CEO at FPD Recycling. “Identification and separation mean that the bromine-free fraction can be recovered.”

Waste recognition

Branding to improve AI logo detection

Despite the superabundance of mixed waste that potentially could be recycled, gaining knowledge of its key constituent statistics is an estimable challenge. “The largest challenge for AI is the sheer variety of waste,” says Peter Hedley at Recycleye.

“Waste differs by country, for instance, and as waste objects can get crushed into strange shapes, get covered in dirt or discolour. AI systems must learn to see past these eventualities to identify the item consistently.”

If initiatives such as Extended Producer Responsibility (a framework that would enfold the environmental costs associated with a product throughout its life cycle into its market price) are to succeed, they need greater transparency as to who – which producer – made it.

A development of EPR that AI-enabled waste management is investigating is in encouraging producers to design packaging so that it is more AI-friendly, i.e. by developing designs that makes it easier for vision systems to recognise specific objects for their recyclability.

Such initiatives are at early stages. Both Recycleye and AMP Robotics are trialling AI/computer vision for brand-level logo detection of retail waste in real-time operation. Recycleye’s algorithm was trained with plastic drinks brands Coca Cola, Pepsi, and Heineken, using 1,399 images from the internet.

An overall accuracy of 93 per cent was obtained with varying lighting conditions, damaged and partially visible objects. Recycleye reports: “With improvement, this technology can be implemented [in MRFs] to support a closed-loop system of re-use. It also provides an Extended Producer Responsibility [for] brands [to] take greater accountability for the environmental impacts of their product-end-life.”

AMP Robotics is working with consumer packaged goods companies to recover higher rates of these materials used in some of their products. One of AMP Robotics’ first initiative partners is beverage brand Keurig Dr Pepper (KDP) in support of its introduction of recyclable K-Cup pods. Following KDP’s conversion of its coffee pods to polypropylene – a valued material for recycled plastics – the companies trained AMP Robotics’ technology to more exactly identify and sort K-Cup pods, with the aim of increasing waste recovery levels.

James Hayeshttps://eandt.theiet.org/rss

E&T News

https://eandt.theiet.org/content/articles/2021/09/ai-driven-robotics-key-to-recycling-s-challenges/

Powered by WPeMatico